🇹🇷 Türkçe: Bu yazının Türkçesini oku →

For two years I have been telling clients the same thing when the board starts asking about AI. The honest answer is not a project plan, it is a tier. Companies are splitting into those with a working answer and those still buying tools and hoping. That split is now visible in the data, and it changes how I would frame the next budget conversation.

Here is what shifted this autumn. BCG’s Applied AI Index for 2026, released 30 September, says nearly half of the 1,330 companies it benchmarked now capture meaningful AI value, up from 5% a year earlier. Gartner, surveying 2,002 CIOs between May and July 2026, says only 13% of CIOs report significant or transformational value. Deloitte, writing from Davos in January, says 25% have moved 40% or more of their pilots into production. Three different numbers measuring three different things, and all three point to the same tier line.

So I went and read what eight named practitioners are saying, on the record, about what sits above that line and what does not. Not to recap their reports. To draw out one question for the 2027 budget cycle: what are companies on the winning side of the split doing that the others are not. The answer is less exciting and more actionable than the hype would suggest.

The Voices on the Record

  • Nicolas De Bellefonds, global AI leader, Boston Consulting Group
  • Jeff Walters, managing director and senior partner, Boston Consulting Group
  • Amanda Luther, managing director and senior partner, Boston Consulting Group
  • Nitin Mittal, Deloitte Global AI leader
  • Jim Rowan, US head of AI, Deloitte
  • John-David Lovelock, Distinguished VP Analyst, Gartner
  • Alicia Mullery, Vice President Analyst, Gartner
  • Aamer Baig, senior partner, McKinsey & Company

The Split Is Now Visible, and It Is Measurable

The headline from BCG’s 30 September 2026 release is that 7.5% of companies qualify as “future-built” and a further 41% are “scaling”. Together, roughly half of the 1,330 CxOs and senior leaders BCG surveyed say they are capturing real value from AI. One year earlier the equivalent number was 5%.

Nicolas De Bellefonds framed the data plainly: “The perception is that AI isn’t delivering,” he said, adding that the picture is more nuanced than that perception suggests. BCG reports that the future-built group shows 2.3 times the total shareholder return of laggards, 2.4 times the revenue growth, and 2.8 times the EBITDA growth.

Nitin Mittal, writing up Deloitte’s 2026 State of AI in the Enterprise from Davos on 21 January, used softer phrasing for the same split. “Across the enterprise, we’re seeing massive ambition around AI,” he said, describing a shift from experimentation to integrating AI into core operations. Deloitte surveyed 3,235 leaders in 24 countries and found that 25% had moved 40% or more of their pilots into production. The other three-quarters have not.

BCG counts value realization, Deloitte counts pilot throughput. Both now show a tier that did not exist a year ago. The question is no longer whether to invest. It is which tier you are in.

People, Not Platforms, Mark the Line

Deloitte’s Jim Rowan put it in one sentence worth keeping on a wall: “The organizations succeeding with AI aren’t just investing in automation and algorithms, they’re investing in their people.” Rowan reports that 25% of leaders now say AI is having a transformative effect on their company, more than double the prior year, and that those companies are the ones pairing tool access with workforce development.

BCG’s Amanda Luther arrived at the same line from a different direction. “AI is reshaping the workforce, not simply shrinking it,” she said in the 30 September release, noting that the pressure falls hardest on middle management. BCG’s index projects a 10% to 15% workforce reduction by 2030 across the surveyed companies and reports that seven in ten future-built and scaling companies are already retraining staff. Dedicated AI roles are projected to rise from 7% of the workforce in 2026 to 22% by 2030.

The common thread is that the winners treat the workforce as a line item in the AI programme, not a cost consequence of it. Training is not a deck at the end of the rollout. It is the rollout. Where I have seen this fall over, the pattern is the same: a platform decision made by IT, a tool licence budgeted without a learning budget, and a workforce that treats the new interface as another thing to keep track of.

Controls That Nobody Built in Time

Jeff Walters, in the same BCG release, flagged the near-term operational risk. “Companies are preparing to hand agents real decision-making authority, and very few have the controls to do so safely,” he said. BCG projects that by 2030, 42% of companies expect to give AI agents genuine autonomy, but only 5% have the full set of required controls today. Agentic AI is expected to account for around 40% of all AI value by 2030.

That is the figure worth sitting with. A decision being made in planning cycles this quarter, to let agents take actions rather than only generate output, is being taken into organisations whose governance was not designed for it. Walters is not saying agentic AI will not happen. He is saying the controls lag the deployment, and the difference between the tiers will include who closed that gap first.

Companies are preparing to hand agents real decision-making authority, and very few have the controls to do so safely.

Jeff Walters, managing director and senior partner, BCG, 30 September 2026

Exit Rules Separate the Vanguard

Gartner’s Alicia Mullery published the sharpest operational finding of the autumn. Presenting the 2027 CIO and Technology Executive Agenda survey at the Orlando symposium on 19 October 2026, she reported that only 31% of CIOs have rules for stopping unsuccessful AI pilots, and only 17% have criteria for abandoning low-value initiatives already in production. The CIOs in the vanguard group, which reported stronger business outcomes, were 2.3 times more likely to have the pilot-stopping rules and 2.6 times more likely to have the production-abandonment criteria.

“Establishing and adhering to these criteria is a team sport,” Mullery said. The phrasing is careful, Gartner’s survey shows an association rather than proof that stopping rules cause the better outcomes, but the direction of the finding is clear. The companies that know when to kill a pilot are the ones generating value from the ones they keep. The companies that cannot kill a pilot cannot free the budget to fund the next one.

Only 13% of the 2,002 CIOs Gartner surveyed said their enterprise had realized significant or transformational AI value. That is narrower than BCG’s 49%, because it measures a higher bar, but the bar it measures is the one a board asks about.

Pruning Beats Piling On

John-David Lovelock, Gartner’s Distinguished VP Analyst on AI spending, put the same lesson in a different register in a 19 May 2026 interview with The National CIO Review. “This is the year that we start pruning the AI garden,” he told the publication. His colleague Daryl Plummer provided the context Lovelock drew on, that 80% of AI projects failed in 2025. Lovelock’s own figure was that 40% of CIOs cannot point to the value AI delivers.

Lovelock is one of the few analysts whose published view this year has been to narrow the AI portfolio deliberately. His argument is that enterprises are moving from sweeping AI overhauls to adding AI to tools they already use. It is the uncomfortable half of Mullery’s finding, said out loud: the way to get into the vanguard is not to add another pilot, it is to delete the ones that are not working and spend the freed capacity on the ones that are.

Scaling Is a Different Craft From Piloting

Aamer Baig, McKinsey’s senior partner in Chicago, wrote the sentence this entire category turns on in May 2024: “It is relatively easy to build gee-whiz gen AI pilots, but turning them into at-scale capabilities is another story.” Two years later, Mullery’s survey is the data behind what Baig called. The honeymoon ended, as he predicted, and the companies in the value tier are the ones that knew the craft of scaling was different from the craft of piloting.

Baig’s second line points at where the energy has to go next: “Getting to scale requires CIOs to focus on fewer things but do them better.” Read alongside Lovelock on pruning and Mullery on exit rules, this is the same argument said three ways. The value tier runs fewer projects, kills the ones that are not pulling, and invests the released capacity in the ones that are.

What This Looks Like on a Mid-Market WooCommerce or WordPress VIP Estate

Every one of these findings has an implementation shape. On a mid-market store or publisher running WordPress or WooCommerce, with somewhere between two and ten thousand SKUs or several hundred content templates, this is what the arguments above cost to ship. Sized for a team of two engineers and a product owner, running alongside normal work.

  • A measurement baseline before the first pilot ships. Two to three weeks of instrumentation: a named KPI per pilot, a 60 to 90-day pre-AI reading in GA4 and server-side logs, a dashboard a non-technical owner can read on a Monday. Without this, Rowan and Mullery are both correct and the pilot will fail the twelve-month review for the same reason every pilot without a baseline fails.
  • Exit rules in writing, in the project tracker, before launch. One day of work, if the arguments have been had. A threshold value, a review date, a named person who presses the stop button. Mullery’s vanguard is 2.3 times more likely to have this in writing; the cost of putting it in writing is a meeting.
  • Centralised AI provider key management. One to two sprints, depending on how many provider accounts are already scattered across environments. A single mu-plugin that reads provider keys from a vault, rate-limits per environment, and logs token spend per feature. Walters is right that controls lag deployment; this closes the smallest and most expensive version of that gap.
  • Governance baked into the content and commerce surface. Two to three sprints. For a publisher, a Gutenberg block pattern that captures provenance (“drafted by”, “reviewed by”, AI involvement, review date) on every AI-assisted post. For a commerce estate, a product data field that records where AI touched the description, feed or attribute set, and whether a human signed it off. This is what Mittal means by integrating AI into core operations.
  • Named owner in the codebase and the org chart. Half a day. A CODEOWNERS entry, a Jira component with a human assigned, a line in the quarterly plan. Luther’s point about workforce reshaping starts with a person who can be invited to the meeting.
  • A quarterly prune. One day per quarter, ongoing. A review meeting that reads the baseline against the threshold and decides which pilots keep going, which get cut, and which get merged. Lovelock’s pruning works only if a calendar event exists for it.

Total standing effort for a mid-market estate: six to eight weeks of concentrated work, plus a measurement quarter, plus the ongoing quarterly prune. That is not a transformation programme. It is the entry fee to the “scaling” tier BCG now measures, and the companies on the other side of the split are doing most of this already, often without naming it.

What None of the Eight Have Resolved

Mullery is explicit that her survey shows association, not causation. BCG’s tier differences are correlations against outcomes the companies self-report. Deloitte’s 25% is a share of pilots moved rather than value captured from them. The honest reading is that the eight voices here agree the split exists, agree on roughly what the vanguard does differently, and disagree on how much of that behaviour is cause and how much is effect of companies that were going to succeed anyway.

The unresolved part is that “what decides which tier you are in” sits on data twelve to eighteen months old at best. Mullery’s survey closed in July 2026. BCG’s index rests on a diagnostic from earlier in the year. We are reading the autumn results of decisions made in 2024 and 2025, which means the actionable variable now is which decisions in this planning cycle are being made with the autumn’s data in mind, and which are being made to the vendor briefing that landed last week.

If you are coming into the budget cycle and the AI line is going up again, these are the questions I would put on the paper before approving it. Which pilot are we cutting this quarter. Who owns the kill rule, by name. What is the baseline the kept pilots are being measured against, and does anyone outside IT know what it is. The vanguard CIOs have already answered them. The 87% who have not yet realized significant AI value are still being asked by their boards. If that conversation is not scheduled, or you want a second pair of eyes on the ones that are, we are the people to talk to.

Leave a Reply

Your email address will not be published. Required fields are marked *

Close Search Window
↑